Chapters the tutor wrote about this player's own games. Other players appear by seat and rank only.
Chapter 1: You keep cheap tenpais dama that LuckyJ declares
You asked whether you stay dama more often than LuckyJ when your hand is closed, in tenpai, riichi is allowed and nobody has declared. You do. Across your 303 Mahjong Soul games you declared 430 of 605 such tenpais. LuckyJ's habits, applied to the same hands, would have declared 466.6. That is 37 riichis fewer, about one every eight games, and 4.8 standard deviations below LuckyJ, too far to be chance.
The missing riichis are cheap hands. With big hands on a poor wait you play like LuckyJ.
The terms
Tenpai: one tile from a winning hand. Closed: no chi, pon or open kan (a closed kan is fine).
Dama: staying in closed tenpai without declaring riichi.
Riichi is allowed when you have 1,000 points or more and at least 4 tiles are left in the wall.
First closed tenpai: the first discard in a hand that leaves you in closed tenpai, with riichi allowed, a wait that is not furiten, and nobody in riichi. Riichi or dama is decided here, so every count below is one per hand.
Han on every winning tile: what the hand is worth on a ron without riichi, on its cheapest winning tile, dora and red fives included. Cheap means no yaku, or 1 or 2 han. No yaku means a ron is impossible without riichi.
Two-sided wait: a ryanmen such as t-i, or a wider wait. Kanchan, penchan, shanpon and single-tile waits are the other waits.
Live tiles: copies of your winning tiles you cannot see anywhere.
LuckyJ's rate: how often LuckyJ declares a hand like yours. It is a logistic fit over LuckyJ's own 2,177 first closed tenpais in its 1,079 Tokujou games: the hand's value and wait, live tiles, turn, dealer, an opponent on two calls or more, and the stage of the game with your place in it. Added over your hands it gives the riichis LuckyJ would have declared. It matches LuckyJ's actual rates by turn, by value and by stage within about a point.
Standard deviation (SD): how far your count sits from LuckyJ's expected count, measured in the chance spread of that count. Under 2 is a maybe; 3 or more is a habit.
Where the riichis went
First closed tenpai
Tenpais
You declared
LuckyJ's rates expect
Gap in SD
No yaku
227
182
196.3
−3.2
Yaku on some winning tiles only
43
42
41.2
+0.6
1 or 2 han
200
153
165.2
−2.9
3 han pinfu, two-sided, 3,900
41
29
34.9
−3.1
Other 3 han or more, two-sided
26
10
11.6
−0.8
3 han or more, other waits
68
14
17.4
−1.0
Cheap hands carry 26 of the 37 missing riichis: 377 of 470 declared against 402.7. The 3-han pinfu, a hand LuckyJ declares 88% of the time (141 of 160), carries 6 more. On a poor wait with 3 han or more you keep the hand dama and so does LuckyJ: 14 of 68 against 17.4, a gap well inside chance.
The gap shows up everywhere in the game. In the East you declared 244 of 319 against 260.7 (−3.2 SD), where your place hardly matters. In South 1 to 3 you declared 160 of 228 against 175.4 (−3.2 SD). In all-last it is 26 of 58 against 30.5, a maybe. From your 13th turn the shortfall is sharper, 27 of 66 against 37.9 (−3.9 SD), but most of the missing riichis come earlier: 403 of 539 against 428.7 through your 12th turn.
When it happened
Your games
Games
Tenpais
You declared
LuckyJ's rates expect
Gap in SD
January to 23 September
87
170
126
135.5
−2.5
The climb, to 1 October
62
132
107
103.9
+0.9
1 October to 16 October
54
106
66
77.4
−3.4
Since 16 October
100
197
131
149.8
−4.2
On the climb you declared like LuckyJ, and the fitted curve sits at +1.6 riichis per 100 tenpais there. After the peak it fell to about −14 by mid-October. The Honver guide's chapter 42 found that gap on 16 October. In the 100 games since, it stayed: the curve averages −9.3, and you declared 131 of 197 against 149.8. Since 16 October the shortfall is in cheap hands, 119 of 154 declared against 132.1, and before your 13th turn, 123 of 180 against 138.6. From your 13th turn there were only 17 such tenpais, 8 declared against 11.2.
Your last 24 games, from 2 November, are back near LuckyJ: 37 of 50 against 38.9. That is too few tenpais to call the habit fixed.
A second opinion from Mortal
Mortal agrees on the direction and sees a third of the size. Over 604 of the 605 tenpais (one replay found no decision), Mortal's riichi probabilities add up to 442 riichis, 12 more than your 430. Taking Mortal's first choice, it would have declared 55 of your 174 dama tenpais and kept 36 of your riichis dama. Mortal declares less often than LuckyJ overall, 73% of these tenpais against 77%. For which hand to declare, LuckyJ's measured habit is the reference here; Mortal is the check on single hands, and on every card below it is quoted.
One hand, turn by turn
16 Jun · East 1 · turn 9 is the first card. East 1, everyone on 25,000, you are West.
Turns 1 to 8. You build a closed hand with no honors left but 6: qe, yi, cc, m,., hjk. On turn 8 you cut m and are one tile from tenpai.
Turn 9. You draw u. Cut 6 and you are tenpai on w, a kanchan. Toimen and shimocha have each cut a w, so 2 are live, and 35 tiles are left. The qe kanchan kills pinfu and the terminals kill tanyao: no yaku. Dama can win only by tsumo, 1,100. Riichi makes a ron possible, 1,300, and lifts a tsumo to 2,000. LuckyJ's rate for this hand is 77% and Mortal's riichi is 88%. You stay dama.
Turns 10 to 13. You draw 4, o, 2 and h and cut each one. The draw you wanted, r for a w-t pinfu, does not come. On its 13th turn kamicha cuts a w; with no yaku you cannot take it. Had you declared on turn 9, kamicha would have been facing the only riichi on the table and might have held that tile, but a riichi gives you the chance and dama does not.
Turn 14. Kamicha declares riichi with 7. You cut your drawn 7, genbutsu.
Turn 15. You draw d and fold with q, genbutsu. Shimocha wins 2,000 from kamicha.
The hand cost you nothing in points, and with 2 live tiles the dama was a small error: LuckyJ keeps about 1 in 3 of these hands dama to look for a better wait. The habit is the problem. Cheap hands with 3 live tiles or more, which LuckyJ declares 71% to 96% of the time before its 13th turn, make up most of your 37.
The cards
Four cards are cheap tenpais you kept dama where LuckyJ and Mortal both declare: the kanchan above; 8 Jul · East 2 · turn 8, no yaku on t-i with 5 live, where LuckyJ declares 95%; 2 Nov · South 1-2 · turn 9, from the games since 16 October; and 19 Sep · South 4 · turn 10, fourth in all-last, where any win was third. Two of those dama hands won anyway, one by tsumo and one after the wait improved. Neither result makes the dama right: the tsumo paid 2,000 where riichi pays 4,000.
Two cards are big hands. 16 Oct · East 4 · turn 5 is a made mangan on a seven-pairs wait: LuckyJ keeps that dama and Mortal declares, so it is close. 10 Oct · East 3-1 · turn 8 is a dealer haneman on a two-sided wait, where riichi adds nothing to a ron. You kept it dama, as LuckyJ does, and won 20,300.
Rule
At your first closed tenpai with nobody in riichi:
No yaku, or 1 or 2 han: declare with 4 live tiles or more at any turn, and with 3 live through your 12th turn. LuckyJ declares these cheap hands 86% of the time on 4 or 5 live and 96% on 6 or more through its 12th turn, 81% and 88% from its 13th, and 71% on 3 live through its 12th. With 2 live or fewer it declares 56%: keep it dama only to change the wait, and declare the turn it improves.
A 3-han pinfu on a two-sided wait: declare. LuckyJ does 88% of the time.
3 han or more on any other wait: stay dama. You already play this like LuckyJ.
All-last: your place decides. From fourth, declare every cheap tenpai (LuckyJ: 35 of 36). From first, keep almost everything dama (LuckyJ declares 5 of 49).
How this was measured
Your 303 game records (January to 10 November 2026, Jade room, South games) against LuckyJ's 1,079 Tokujou games, first closed tenpais as defined above, nobody in riichi.
scripts/contrast/tenpai.py data/self_games/majsoul/index.json tenpai_player.jsonl --procs 4, and the same over LuckyJ's manifest filtered to mode == "tenhou-tokujou", giving tenpai_luckyj_tokujou.jsonl.
scripts/ch01_dama_check.py tenpai_player.jsonl tenpai_luckyj_tokujou.jsonl ch01_dama_check.json: LuckyJ's logistic fit (the Honver guide's chapter 33 terms from mine_climb_check.py, plus stage and place: South 1 to 3 or all-last, by 1st to 4th), its calibration, every table above, and the card candidates ranked by LuckyJ's rate. The SD is (declared − expected) over the square root of the summed binomial variances, without the fit's own uncertainty.
scripts/ch01_mortal_dama.py ch01_dama_check.json OUT.json --key dama and --key declared: Mortal's probability on riichi at each first tenpai, from a replay of the whole game with the Mortal policy.
scripts/ch01_figure.py tenpai_player.jsonl tenpai_luckyj_tokujou.jsonl fig-riichi-vs-luckyj.png: one point per game in date order, a local linear fit with a 20-game Gaussian kernel over the tenpai-level residuals, and a 95% band from 1,000 bootstrap resamples of games.
Cards: scripts/build_personal_guide.py --spots spots.json --out guide-01-dama-habit.json --ja-out guide-01-dama-habit-ja.json; every recorded cut replays. Points come from the mahjong library's hand calculator with Mahjong Soul's red fives.
Hands from your games
A kanchan with no yaku and 2 live tiles, kept dama
16 Jun · East 1 · turn 9
The spot
East 1, turn 9, 35 tiles left in the wall. Everyone has 25,000; you are West and toimen is the dealer. You draw u into tenpai: cut 6 and you wait on w with qe. Toimen and shimocha have each cut a w, so 2 are live. The hand has no yaku. Nobody has called or declared.
What you did
You cut 6 and stayed dama. Without a yaku a ron was impossible, so only a tsumo could win, for 1,100. You stayed dama for six turns; kamicha cut a w on its 13th turn that you could not take, and on turn 15 you broke the tenpai. Shimocha won the hand.
LuckyJ and Mortal
LuckyJ declares 85% of its first tenpais with no yaku (666 of 780). With 2 live tiles or fewer it declares 57% (28 of 49), and 68% through its 10th turn. Its fitted rate for this hand is 77%. Mortal puts 88% on riichi.
The fix
Declare. Riichi is the hand's only yaku: it makes a ron possible at all, 1,300, and lifts a tsumo from 1,100 to 2,000. With two live tiles this is a lean: LuckyJ keeps about 1 in 3 of these dama. If you keep one, keep it to change the wait, and declare the turn it improves.
No yaku on a two-sided wait with 5 live, kept dama
8 Jul · East 2 · turn 8
The spot
East 2, turn 8, 41 tiles left. You are second on 24,500; kamicha is the dealer on 27,000. You draw l into tenpai: cut 2 and you wait on t-i, 5 tiles live. No yaku: the v triplet rules out pinfu and the terminals rule out tanyao. Toimen has two calls, a 4 pon on its own seat wind and a bnm chi; shimocha has ponned 5. Nobody has declared.
What you did
You cut 2, genbutsu against both callers, and stayed dama. Four turns later you won by tsumo on a red t for 2,000.
LuckyJ and Mortal
With no yaku and 5 or more live tiles LuckyJ declares 95.5% of its first tenpais (276 of 289), and 90% (54 of 60) when an opponent already shows two calls. Its fitted rate for this hand is 95%. Mortal puts 99.6% on riichi.
The fix
Declare. Dama could win only by tsumo. With riichi the same red-five tsumo pays 4,000 instead of 2,000, and a t or i from any of the three others wins too. Two open hands across the table are a reason to hold a safe tile, not a reason to give up the ron.
Third in South 1, a kanchan with no yaku, kept dama
2 Nov · South 1-2 · turn 9
The spot
South 1, two honba, turn 9, 34 tiles left. You are third on 13,900, 2,000 above shimocha in fourth and 22,000 behind kamicha in second. Toimen is the dealer on 38,300 and has ponned 4. You draw a red t into tenpai: cut r and you wait on k with jl, 3 tiles live. No yaku: the o and l rule out tanyao. Nobody has declared.
What you did
You cut r and stayed dama, then cut a and . from the wall. Kamicha won by tsumo, and you paid 700.
LuckyJ and Mortal
With no yaku and 3 live tiles LuckyJ declares 76% of its first tenpais (127 of 168), and 78% in South 1 to 3 (46 of 59). Its fitted rate for this hand, third in South 1, is 78%. Mortal puts 99.6% on riichi.
The fix
Declare. Dama wins only by tsumo, 2,000 with the red five. Riichi makes that tsumo 4,000 and adds a 2,600 ron. The stick puts you on 12,900, 1,000 above fourth, and that is the real cost here; both LuckyJ and Mortal pay it.
Fourth in all-last, any win is third, kept dama
19 Sep · South 4 · turn 10
The spot
South 4, the last hand, turn 10, 32 tiles left. You are fourth on 5,200, 2,000 behind shimocha. Kamicha is the dealer on 61,900 and toimen is second on 25,700. You draw t into tenpai: cut i and you wait on g with fh, 3 tiles live. The hand holds the r dora and a red b but no yaku. Shimocha and toimen have each called once; nobody has declared.
What you did
You cut i and stayed dama. The next turn you drew j, switched to g-k for pinfu, and later won g from kamicha for 3,900, which took you to third.
LuckyJ and Mortal
In all-last from fourth, LuckyJ declares 97% of its cheap first tenpais, no yaku or 1 or 2 han (35 of 36). Its fitted rate for this hand is 93%. Mortal puts 99.6% on riichi.
The fix
Declare, though this one ended well. Any win lifts you to third, and at Saint 3 a fourth costs about 250 rank points more than a third. Dama could get there only by tsumo, 4,000. Riichi wins on any g: 5,200 by ron from any of the three, 7,900 by tsumo. The j that gave you pinfu came the next turn, and you cannot count on that.
A made mangan on a seven-pairs wait, kept dama
16 Oct · East 4 · turn 5
The spot
East 4, turn 5, 52 tiles left. You are second on 26,000; kamicha is the dealer on 29,600. You draw m into seven-pairs tenpai. Cut 6 and you wait on n, the dora, with 3 live; cut n and you wait on 6 with 2. With n the hand is seven pairs, tanyao and two dora: 8,000 dama by ron, and a tsumo is a haneman, 12,000, already. Nobody has declared.
What you did
You cut 6 and stayed dama. Shimocha declared riichi on the next discard. You kept the dama and dealt in to shimocha on turn 13 with w for 5,200.
LuckyJ and Mortal
On a wait that is not two-sided, LuckyJ keeps hands of 5 han or more dama: it declared 5 of 43 (12%), and 4 of 25 through its 8th turn. Its fitted rate for this hand is 11%. Mortal disagrees and puts 99.5% on riichi.
Verdict
Close call
A dealer haneman on a two-sided wait, kept dama
10 Oct · East 3-1 · turn 8
The spot
East 3, one honba, turn 8, 41 tiles left. You are the dealer and first on 26,000; the others sit between 24,000 and 25,000. You draw u into tenpai: cut x and you wait on e-y, 6 live. Tanyao, three , as dora and two red fives: 18,000 on a e ron, and 24,000 on y with the 4-5-6 triple sequence. Nobody has declared.
What you did
You cut x and stayed dama. Toimen declared riichi later; you held the dama and won by tsumo on e at turn 16, 6,000 all, 20,300 with the honba and both sticks.
LuckyJ and Mortal
With 5 han or more on a two-sided wait LuckyJ declares 27% of its first tenpais (8 of 30), and 1 of 7 as the dealer. Its fitted rate for this hand is 32%. Mortal puts 2% on riichi.